Professional Documents
Culture Documents
Spinors are analogous to vectors and tensors, but they transform according to the “cov-
ering group” SL(2, C) of the homogeneous Lorentz group rather than the Lorentz group
itself. SL(2, C) will be discussed in some detail later—for now we just mention that it acts
most directly on two-component, complex vectors, the basic spinors that are analogous to
four-vectors for the Lorentz group. The aim of these notes is not particularly to develop
the so-called “spinor calculus,” which is a way of describing the finite dimensional repre-
sentations of the Lorentz group, but rather to develop some basic facts about the groups
themselves.
After some thought about whether it is really appropriate for students approaching
Lorentz spinors for the first time to see the complex Lorentz group, we have decided to
include it here. The need to actually deal with complex Lorentz transformations occurs
only in rather technical situations, especially involving questions of analyticity or continu-
ation from the Minkowski to the Euclidean domain; but the cost of including them in the
discussion is so minimal that there seems no reason to deny the potential benefit of the
richer context. It is a recurring theme in symmetry considerations where a given group
sits relative to its subgroups, and to the groups that naturally contain it as a subgroup.
But the essential results for relativistic physics are indeed those for the real Lorentz group
and its “covering group” SL(2, C).
The real, homogeneous Lorentz group L(IR) is the set of 4×4 real matrices that satisfy
1
the equation
1 0 0 0
0 −1 0 0
Λ Tr
GΛ = G, G = . (1.1)
0 0
−1 0
0 0 0 −1
It is easy to check that this does define a group. Our notation is µ = 0, 1, 2, 3 for Minkowski
indices (latter part of the Greek alphabet); and the Minkowski metric gµν = g µν is given
by the matrix elements of G. In tensor notation, we have
0 0
Λµµ0 Λν ν 0 g µ ν = g µν , (1.2)
where Λµν is the “natural” notation for the matrix elements of Λ, corresponding to a
contravariant vector transformation law such as
0
x0 = Λx , x0µ = Λµν xν , x ∈ IR4 ; (1.3)
?
and g µν and gµν are equally natural notations for the matrix elements of G.
Since we have mentioned contravariant vectors, recall that the metric symbol gµν can
be used to lower the Minkowski index of a contravariant vector and turn it into a covariant
vector (like the gradient ∂/∂xµ):
0
xµ = gµν xν , x0µ = Λµν xν 0 , (1.4)
where we have used the fairly common, but we think bad notation,
0 0
Λµν = gµµ0 g νν Λµν 0 , (1.5)
†
for the matrix elements of the contragredient matrix
Λ−1T r = G Λ G . (1.6)
While it may seem economical, the notation Λµν for the matrix elements of the contra-
gredient matrix encourages the misconception that Λ is a tensor. It is better to write the
?
Unless otherwise stated, when we write x ∈ IR4 for a four-vector, we take it to be contravariant.
†
The contragredient of a matrix is its inverse transpose (or transpose inverse). It has the special form
GΛG here because of Eq. (1.1).
2
covariant transformation law as
x0µ = xν Λν µ , (1.7)
or to make it explicit that Λ−1T r is a different matrix whose matrix elements are naturally
written with down and up indices by giving it a name,
The parenthesis notation for the indices on the r.h.s. in the second equation is handy for
some calculations which are easier to follow in matrix than in index notation.
det Λ = ±1 , (1.9)
and that
X
3
Λ0 0 Λ0 0 − Λ0 i Λ0 i = 1 , |Λ0 0 | ≥ 1 . (1.10)
i=1
Name det Λ Λ0 0
L↑+ (proper, orthochronous) +1 ≥ 1
L↑− (improper, orthochronous) −1 ≥ 1
L↓+ (proper, nonorthochronous) +1 ≤ −1
L↓− (proper, orthochronous) −1 ≤ −1
P = G, T = −G , Y = −I , (1.11)
show that none of the four categories is empty; and in fact the sets of transformations are
related as follows:
L↑− = L↑+ P = P L↑+ , L↓+ = L↑+ Y = Y L↑+ , L↓+ = L↑+ T = T L↑+ . (1.12)
3
other three pieces are continuously connected to the matrices P , Y , and T , respectively,
and from the defining properties of the four pieces listed in the table, it is clear that all
are disconnected from each other.
The set of complex, 4×4 matrices that satisfy Eq. (1.1) is also a group, called the
complex, homogeneous Lorentz group, L(C).
Its elements also satisfy det Λ = ±1, but the orthochronicity conditions no longer
make sense for all Λ. The group L(C) has two connected pieces, differing by space or time
inversion from each other:
Of the two pieces, only L+ (C) is a subgroup; it is the identity component of L(C). The
matrices I and −I are connected in L+ (C), and P and T are connected in L− (C).
The equation
Λ† G Λ = G , (1.14)
?
where “† ” means Hermitean conjugation, defines another complex Lorentz group, but
it is the wrong one for considerations of analyticity. As defined at the beginning of this
†
section, L(C) is the analytic complexification of L↑+ , a technical notion which means that
L+ (C) is a complex analytic manifold (of complex dimension six), which contains L ↑+ as
a real analytic submanifold (of real dimension six). L+ (C) is the only group extension of
L↑+ with that property.
It is known from undergraduate physics that L(IR) is the group of real linear transfor-
?
The Hermitean conjugate of a matrix is its complex conjugate transpose.
†
A classic reference on group manifolds is L. Pontrjagin, Topological Groups, Princeton University Press,
1939.
4
mations of IR4 that leave the Minkowski scalar product invariant:
It is also true that L(C) is the group of linear transformations of C4 that leave the complex
Minkowski scalar product invariant. The complex scalar product is defined by
z · z = (z 0 )2 − z · z , (1.16)
The proper three-dimensional group of real rotations, which is the proper orthogonal
to a subgroup of L↑+ , namely, the set of real, 4×4 matrices
?
group O+ (3, IR), is isomorphic
of the form
1 0 0 0
Λ=
0 ,
RT r R = I .
R
det R = 1 , (1.17)
0
0
When it doesn’t seem confusing, we may also use the notation R for the 4×4 matrix to
which it corresponds.
If we let R be complex, but still orthogonal and with determinant unity, we get the
group of proper complex rotations of C3 , called O+ (3, C), as a connected subgroup of
L+ (C).
?
A group isomorphism is a one-to-one map of one group onto another which preserves the group mul-
tiplication law. A homomorphism is a possibly several-to-one map of one group into another that
preserves the group law. A homomorphism of a group into another group is also called a representation
of the group, especially when it is into a group of matrices. A representation is called faithful when it
is one-to-one. A group automorphism is a group isomorphism onto the group itself.
5
1.2 Matrix Representation of Minkowski Space
The set of complex four-vectors C4 , with the Minkowski metric, can be put in one-to-
one linear correspondence with the set of complex 2×2 matrices, since both have the same
complex dimension. To make the correspondence explicit, we introduce a complete set of
2×2 matrices, the Pauli matrices:
σµ = (I, −
→
σ ), (1.18)
! ! ! !
1 0 0 1 0 −i 1 0
σ0 = , σ1 = , σ2 = , σ3 = . (1.19)
0 1 1 0 i 0 0 −1
The Pauli Matrices are naturally defined as a covariant vector. An alternative complete
set is
σ̃µ = (I, −−
→
σ ). (1.20)
Z = z · σ ≡ z µ σµ = z 0 I + z · −
→
σ . (1.21)
These linear correspondences are one-to-one, onto inverses of each other, because of the
orthogonality of the Pauli matrices:
Tr σi = 0 . (1.25)
6
σi σj = δij I + i ijk σk . (1.27)
?
The real four-vectors x ∈ IR4 are put in one-to-one correspondence with the Hermitean
2×2 matrices:
X = x · σ = x · σ† = X † (1.28)
σµ† = σµ . (1.29)
The Minkowski scalar product goes over into the determinant in 2×2 space:
z · z = det(z · σ) , (1.30)
Finally, the reader should check some useful identities, good for any a, b, c ∈ C4 :
a · σ b · σ̃ + b · σ a · σ̃ = 2 a · b , (1.32)
a · σ a · σ̃ = a · σ̃ a · σ = 2 a · a , (1.33)
a · σ b · σ̃ c · σ = a · σ b · c − b · σ c · a + c · σ a · b + i µνλρ σµ aν bλ cρ , (1.34)
where 0123 = −1. Note that Eq. (1.34) has another version where σ and σ̃ are interchanged
and i is replaced by −i, which can be derived by evaluating Eq. (1.34) at P a, P b, and P c,
and applying the definitions (1.20) of σ̃ and (1.11) of P , and the pseudoscalar nature of
the Levi-Civita alternating symbol:
0 0 0 0
Λµµ0 Λν ν 0 Λλλ0 Λρ ρ0 µ ν λ ρ = det Λ µνλρ . (1.35)
This formula holds, by the way, for any complex 4×4 matrix Λ.
?
A Hermitean matrix is equal to its Hermitean conjugate.
7
1.3 Covering Groups of L↑+ and L+ ( C)
• The first is that the connected subgroups of the real and complex homogeneous
groups are multiply connected, and the covering groups are constructed as sim-
ply connected. The covering groups are distinct from the groups they “cover,” not
isomorphic, but homomorphic to them. In our case the homomorphism from the
covering group to the group is two-to-one.
• The second is that the Lie algebra of infinitesimal generators of a Lie group is iso-
morphic to the Lie algebra of infinitesimal generators of its universal covering group.
An equivalent technical statement is that a Lie group is locally isomorphic to its
universal covering group. The Lie algebra is an essential focus for physics, because
the infinitesimal generators are observables, and the algebraic relations they obey
are an essential part of their physical meaning in quantum mechanics.
For physicists, a rough but useful characterization is that the Lorentz group has to do
with the symmetry of vectors (tensors), while its covering group has to do with that of
spinors.
Consider the linear transformations of the space of 2×2 matrices that have the form
Z0 = A Z BT r , (1.36)
where A and B are 2×2 matrices. We restrict ourselves to the class of pairs (A, B) that
preserve the Minkowski metric; i.e.,
Clearly the pair (cA, B/c) gives the same linear transformation of 2×2 matrices as (A, B),
if c is any nonzero complex number. Equation (1.38) tells us that we can choose c to
8
arrange
In other words, we need only consider A and B that are unimodular (determinant
unity), if we want the pair (A, B) to preserve the Minkowski metric. From now on, we
do that. There is still the freedom of a sign; (A, B) and (−A, −B) give the same metric-
preserving transformation; and det A = det(−A) = det B = det(−B) = 1.
It is just the direct product group SL(2, C)×SL(2, C), where SL(2, C) denotes the group of
2×2 matrices with unit determinant (special linear group of transformations in 2 complex
dimensions).
Since the correspondence we defined between four-vectors and 2×2 matrices is linear,
the linear transformation (A, B) of 2×2 matrices induces a linear transformation Λ(A, B)
of four-vectors. That correspondence is easy to compute; let
z0 · σ = A z · σ B T r
(1.41)
= [Λ(A, B) z] · σ .
From the orthogonality relation for Pauli matrices (1.23), it follows that
Because Λ(A, B) preserves the Minkowski metric, it must belong to L(C). Moreover,
it is easy to show from Eq. (1.41) that
What we have done, then, is to construct a homomorphism of the group SL(2, C)×SL(2, C)
into L(C). One of our jobs is to find out which Lorentz transformations we have arrived
9
at in this way. We are going to see that we don’t get them all. But before we do that, let’s
find out how many transformations (A, B) correspond to a given Λ. The homomorphism
is at least two-to-one, because
We claim that the homomorphism is not more than two-to-one, by the following argu-
ment. Suppose two pairs, (A, B) and (A0 , B 0 ) induce the same Lorentz transformation:
A σµ B T r = A0 σµ B 0T r . (1.46)
σµ = A−1 A0 σµ B 0T r B T r−1
(1.47)
= C σµ D T r , C ≡ A−1 A0 , D ≡ B −1 B 0 .
C = λ1 I , D = λ2 I , (1.48)
then we have det C = det D = 1 = λ21 = λ22 , so that λ1 = ±1 and λ2 = ±1. Also, we
cannot have λ1 λ2 = −1, for that contradicts (1.47). We then have C = D = ±I, which
proves that (A0 , B 0 ) = (A, B) or (−A, −B), as claimed.
Proof. Because the trace is linear, it is sufficient to prove the lemma for each of the four
10
Pauli matrices. Consider M = σ0 = I. Applying the identity (1.32), we get
σ̃µ σ0 σ µ = σ̃µ σ µ = σµ σ̃ µ
= 1/2 g µν (σµ σ̃ν + σν σ̃µ ) (1.50)
= 1/2 g µν 2 gµν I = 4I = 2 (Tr σ0 ) I .
Next, consider M = σi , i = 1, 2, 3:
σ̃µ σi σ µ = σ0 σi σ 0 + σj σi σj
= σi + δji σj + i jik σk σj (1.51)
= 2 σi − 2 σi = 0 = 2 (Tr σi ) I ,
Now apply the lemma to (1.47), after multiplying from the left or right by σ̃ µ and
contracting. For example,
σ̃ µ σµ = σ̃ µ C σµ DT r ,
(1.52)
4 I = 2 (Tr C) DT r .
This shows that D is a multiple of I, and a similar argument works for C.
The reader should check that the lemma above can be used to prove the following
refinement of the statements in Eqs. (1.47) and (1.48):
σµ M = N σ µ
for all four Pauli matrices. Then M = N and both are the same multiple of the identity.
This theorem contains the statement that the three Pauli matrices σi are irreducible;
namely, any 2×2 matrix that commutes with all three is a multiple of the identity.
Λ(A,B)
SL(2, C)×SL(2, C) −−−−−−→ L(C) .
The next claim is that Λ(A, B) is in fact a proper Lorentz transformation. We give a
heuristic argument; namely, you have to believe that SL(2, C) is connected. We shall see
11
that explicitly later; but for now we point out that SL(2, C) is the set of matrices A = a · σ
with det A = a · a = 1; and that the complex hyperboloid a · a = 1 is a connected set in
C4 . It follows that SL(2, C)×SL(2, C) is also connected. Note that Λ(I, I) = I. Since
?
Λ(A, B) as defined in (1.43) is a continuous function of A and B, and any pair (A, B)
is continuously connected to (I, I), we see that Λ(A, B) is continuously connected to the
identity in L(C). Therefore det Λ = 1, for det Λ is a continuous function of Λ, and there is
no continuous way to reach group elements with det Λ = −1 from the identity.
That shows that Λ(A, B) ∈ L+ (C). A stronger statement is true: the homomorphism
Λ(A, B) maps SL(2, C)×SL(2, C) two-to-one onto L+ (C). The proof that every Λ ∈ L+ (C)
has the representation Λ = Λ(A, B) is algebraically nontrivial, and we omit it. One
could do it by inverting the equation to find A and B, given Λ, as Joos does in his
?
famous article; or one could do it by parametrizing L+ (C) in terms of a product of six
elementary transformations in the six µν planes, and putting that in correspondence with
a parametrization of SL(2, C)×SL(2, C).
Now we want to find out which pairs (A, B) correspond to the most important sub-
groups of L+ (C). First, let’s consider the proper complex rotations, O+ (3, C). The answer
is easy; O+ (3, C) is the set of proper Lorentz transformations that leave the µ = 0 com-
ponent of any four-vector unchanged, so
A σ0 B T r = σ 0 =⇒ B T r = A−1 . (1.53)
Thus
n o
−1T r
O+ (3, C) = Λ(A, A ) : A ∈ SL(2, C) . (1.54)
To find those pairs (A, B) which give real Lorentz transformations, we use the fact
?
That is, of their matrix elements.
?
H. Joos, Fortschritte der Physik 11 (1962) 65.
12
that real transformations preserve real four-vectors; and hence for any Hermitean X,
(A X B T r )† = B T r† X A† = A X B T r ,
(1.55)
X = A−1 B T r† X A† B T r−1 .
In particular, this must hold for X = σµ ; and we have already shown in Section 1.3.c that
any 2×2 matrix that commutes with all four σµ ’s must be a multiple of the identity. More
precisely in this case:
A−1 B T r† = B −1 A†T r = ±I ,
(1.56)
A = ±B T r† = ±B̄ .
Thus, the pairs that give real Lorentz transformations fall into two classes, those of
the form (A, Ā) and those of the form (A, −Ā). Each class is connected, because SL(2, C)
is. The first is connected to the identity (I, I), which is outside the second class; so the
two classes are disconnected from each other.
It is correct to guess that these two sets correspond to L↑+ and L↓+ , and we can guess
at once that (A, Ā) goes with L↑+ while (A, −Ā) goes with L↓+ , because the (A, Ā) are a
subgroup of SL(2, C)×SL(2, C) (isomorphic to SL(2, C) itself), while the (A, −Ā) are not,
corresponding to the fact that L↑+ is a subgroup of L+ (C), while L↓+ is not a group.
To verify that, note that if X represents a positive, timelike vector, then X is non-
negative definite. That is so because X, being Hermitean, has eigenvalues λ1 and λ2 ;
and
λ1 + λ2 = Tr X = 2 x0 ≥ 0 ,
(1.57)
λ1 λ2 = det X = x · x ≥ 0 ,
implies that λ1 and λ2 are nonnegative, after a short calculation. Therefore AXA† is
nonnegative definite, too; and
while
This is equivalent to saying that Λ(A, Ā) is orthochronous, and that Λ(A, −Ā) is nonortho-
13
chronous. Indeed
Λ(I, −I) = Y , (easy to show)
(1.60)
Λ(A, −Ā) = Λ(A, Ā) Λ(I, −I) = Λ(A, Ā) Y ,
We conclude that Λ(A, Ā) defines a two-to-one homomorphism of SL(2, C) onto L↑+ .
That the homomorphism is onto follows from the analysis above if we grant that Λ(A, B)
is onto L+ (C). We introduce the notation
(Λx) · σ = A x · σ A† . (1.62)
Λ(A)
SL(2, C) −−−−−→ L↑+ .
The subgroup of L↑+ that leaves x0 invariant for all real four-vectors x ∈ IR4 is O+ (3, IR),
and it corresponds to those (A, Ā) satisfying
A σ0 A † = σ 0 , A A† = I . (1.63)
That is, A is unitary and unimodular, A = U ∈ SU2 . The subgroup SU2 of SL(2, C) is
thus two-to-one homomorphic to O+ (3, IR), and it is known to be its covering group.
From now on the focus will be on real Lorentz transformations; and where confusion
seems unlikely, we write simply Λ = Λ(A). In terms of SL(2, C), the (real) Lorentz
14
transformation reads
(Λx) · σ = A x · σ A† ; (1.64)
(Rx) · −
→
σ = U x·−
→
σ U† , U ∈ SU2 . (1.65)
Most of the discussion so far has been based on the correspondence x ↔ x · σ. It should
be realized that we could have developed the theory via the alternative correspondence
x ↔ x · σ̃. We should then have gotten another homomorphism of SL(2, C) onto L↑+ —let’s
?
call it Λ̃(A):
[Λ̃(A) x] · σ̃ = A x · σ̃ A† . (1.66)
To find how Λ and Λ̃ are related to each other, recall (1.33) that
x·σ
(x · σ̃)−1 = . (1.67)
x·x
Taking the inverse of both sides of (1.64), we get
Hence
We can show besides that Λ̃ = GΛG, because from (1.66) we can write
where Λ = Λ(A). When put this together with (1.62), it is evident that combinations such
?
Although Λ̃ turns out to be the same as defined in Section 1.1.a, the two definitions should be treated
as independent for the moment.
15
as x · σ y · σ̃ and x · σ y · σ̃ z · σ, etc., are covariant if the correct SL(2, C) transformations
are supplied; e.g.,
Λx · σ Λy · σ̃ = A x · σ y · σ̃ A−1 ,
Λx · σ Λy · σ̃ Λz · σ = A x · σ y · σ̃ z · σ A† ,
(1.73)
Λx · σ̃ Λy · σ = A†−1 x · σ̃ y · σ A† ,
Λx · σ̃ Λy · σ Λz · σ̃ = A†−1 x · σ̃ y · σ z · σ̃ A−1 .
Although the representations Λ and Λ̃ of L↑+ are similar to each other via G, the
corresponding respresentations A and A†−1 of SL(2, C) are not similar. That is, there is
no 2×2 matrix M such that A†−1 = M AM −1 for all A ∈ SL(2, C). For if there were, we
would have
M = A† M A ;
m = Λ(A† ) m
for all A† , and hence for all Λ ∈ L↑+ . It is a basic fact that the only four-vector with that
?
property is m = 0, so M = 0, which is a contradiction.
16
Lemma. Let
!
0 1
ε ≡ i σ2 = . (1.74)
−1 0
Then
Proof. Write down the four equations explicitly, and use the properties of the Pauli
matrices (1.20) and (1.24).
ε M ε−1 = m · σ̃ T r
= [(m · σ)−1 m · m]T r (1.77)
= M −1T r det M ,
Even when det M = 0, it is a fact that the statement in the theorem remains both well
?
defined and true.
17
1.3.g Discrete Transformations
The elements P , T , and Y of L(IR) form a discrete subgroup, each member of which
is its own inverse. They can also be applied as similarity transformations on elements of
L↑+ :
P Λ P −1 = T Λ T −1 = Λ̃ ,
(1.79)
Y Λ Y −1 = Λ ,
which in fact map L↑+ onto itself in a one-to-one fashion and preserve the group multipli-
cation law. In other words, they induce automorphisms of L↑+ , with P and T both giving
the same thing, the contragredient (because P = −T = G), and Y giving the identity
automorphism.
On the other hand, we saw a correspondence in Section 1.3.f between the automor-
phisms A 7→ A†−1 of SL(2, C) and Λ 7→ Λ̃ of L↑+ . We should ask what this has to do with
extending the homomorphism between SL(2, C) and L↑+ to include the discrete symmetries;
in other words, what is the spinor representation of P , T , and Y ?
The answer is not as simple as one might hope. The results in Section 1.3.f can also
be used to show that there is no 2×2 matrix representation of P or T . If there were
such a matrix MP for P , for example, it would have to induce a similarity transformation
MP AMP−1 = A†−1 ; and we have found that to be impossible. The situation changes
when one considers the four-dimensional Dirac representation of SL(2, C), which extends
naturally to include the discrete transformations as matrices.
18